Free Subgroups of the Fundamental Group of the Hawaiian Earring

Journal of Algebra 219, 598 605 (1999) Article ID jabr.1999.7912, available online at http://www.idealibrary.com on Free Subgroups of the Fundamental Group of the Hawaiian Earring Katsuya Eda School of Science and Engineering, Waseda University, Tokyo 169-0072, Japan E-mail: eda@logic.info.waseda.ac.jp Communicated by Walter Feit Received October 26, 1998 Generalizing Specker s result [7] for a countable case without using the continuum hypothesis, Nöbeling [6] proved that the subgroup of a direct product I consisting of all finite valued functions is free for any index set I As a non-commutative version of this theorem in case I is countable, Zastrow [8] proved that the subgroup of consisting of elements which have certain restricted presentations is free, where is isomorphic to the fundamental group of the Hawaiian earring [3]. His proof is involved with complicated notions related to the inverse limit of free groups. Here, we use the notion words of infinite length from [3, 4] and give a simplified proof. Since we do not depend on countability in the proof, we prove Zastrow s theorem for the free complete product I for any index set I On the other hand, Cannon and Conner [1] proved that another subgroup of is free. They used the notion words of infinite length, but did not use a simple reduction for a finite concatenation of reduced words (see Lemma 1.7). We give a short proof of their theorem. We remark that the group I is called a big free group in [1]. Advantages of using words are the existence of reduced words and that of a simple reduction based on Proposition 1.3, which were used to investigate group theoretic properties in [3, 4, 2]. In Section 1, we state definitions and preliminary facts about words of infinite length, including the above facts about reduced words. In Section 2, we state the main theorem precisely and prove it. In the Appendix, we explain relationships between loops in the Hawaiian earring and the above two free subgroups of are free. 0021-8693/99 $30.00 Copyright 1999 by Academic Press All rights of reproduction in any form reserved. 598

free subgroups 599 1. DEFINITIONS AND PRELIMINARY FACTS Since we are concerned with the infinitary version of free groups, we use infinitary words of special type, which were used in [4, Sect. 3]. Definition 1.1. For an index set I a letter is an element of the set i i i I A word W I is a function from a linearly ordered set W to i i i I W 1 = α 1 α W denotes the inversely ordered set of W where α 1 <β 1 if β<α W 1 is a word such that W 1 = W 1 and W 1 α 1 = W α We identify two words V and W if there exists an order preserving bijection ϕ V W such that V α =W ϕ α for each α V for which we write V = W Let i be a copy of the integer group and let p FG i G i i F i be the projection for finite subsets F G of I The notation F I means that F is a finite subset of I Thus, the unrestricted free product is lim i F i p FG F G I [5]. For a word W I and F I W F is the word obtained by deleting all letters ±i i I \ F from W i.e., W F = α W W α =±i for some i F and W F = W W F W F can be regarded as an element of the free product i F i Two words V and W are equivalent, if V F = W F as elements of the free product for any F I The equivalence class containing W is denoted by W Then, we get a free complete product i I i = W W I where the operation is defined by the concatenation. This group is isomorphic to the subgroup F I i F i lim i X i p XY X Y I\F of the unrestricted free product [3, Proposition 1.8]. Definition 1.2. A word V is a subword of W if there are words X and Y such that W = XVY V is a proper subword in the case at least one of X or Y is not empty. V is an initial subword in the case X is empty. A word is reduced if there exists no non-empty subword which is equivalent to the empty word. Though the definition of reducedness in this paper is different from that of [3, Definition 1.3], the difference corresponds to that between a free group and a free product of groups, and so it is not essential. Therefore, we also state propositions without proofs, which correspond to the indicated propositions in [3]. Proposition 1.3 [3, Theorem 1.4]. For each word W I there exists a unique reduced word V I such that V = W Proposition 1.4 [3, Corollary 1.7]. Let V and W be reduced words. Then there exist reduced words X Y Z such that V = XY and W = Y 1 Z hold and XZ is reduced.

600 katsuya eda Definition 1.5. For i I and a word W I l i W denotes the number of appearances of ±i in W i.e., the cardinality of α W W α = ±i A word W is of appearance n if max l i W i I =n The subgroup of bounded appearance of i I i is Bd = W W is of appearance n, n<ω Let Bd n be the subgroup of Bd generated by words of appearance less than or equal to n A subgroup G of i I i is fine if the following holds: In the case V is a subword of a reduced word W and W G holds, V G also holds. For a subset X of a group G X denotes the subgroup of G generated by X If no confusion is possible, we shall follow the convention and use W to denote the group element W Lemma 1.6. Let G be a fine subgroup of i I i and let W be a reduced word. Then G V an initial subwords of W is fine. Proof. Let X 0 X m be reduced words which belong to G or initial subwords of W It can be easily proved by induction on m that any initial subword of the reduced word of X 0 X m belongs to G V an initial subword of W Since any subword is a product of the inverse of an initial subword and another initial subword, we get the conclusion. Lemma 1.7. Let A 0 A n and W be reduced words and let W = A 0 A n Then there exist reduced words W 0 W m and 0 i 0 < < i m n satisfying the following: W = W 0 W m and each W k is a subword of A ik. A 0 A n = X0 W 0 X m W m X m+1 for some words X i such that X i = e for each i Proof. The proof is by induction on n Since the case n = 0 follows from Proposition 1.3, we prove the induction steps. Let A be the reduced word for A 0 A n 1 Then, W = AA n holds and hence there exist reduced words X B, and C such that A = BX A n = X 1 C, and W = BC by Proposition 1.4. Apply the induction hypothesis to A and let V 0 V k be the obtained reduced words. Now, B = V 0 V i 1 V i where V i is an initial subword of V i Hence, W = V 0 V i 1 V i C holds and V 0 V i 1 V i C satisfy the required properties. 2. MAIN THEOREMS AND THEIR PROOFS In this section, we prove the following theorems, the first of which was proved by Zastrow [8] in the case I is countable and the second of which was proved by Cannon and Conner [1].

free subgroups 601 Theorem 2.1 (The countable case is due to Zastrow [8]). Bd of bounded appearance of i I i is free. The subgroup To state the next theorem, we recall scattered sets. For a space X X denotes the subspace of X consisting of all non-isolated points. Let X 0 = X and X α+1 = X α for an ordinal α and let X α = X β β<α for a limit ordinal α A space is said to be scattered if X α is empty for some α Since a linearly ordered set can be regarded as topological space with its order topology, we call a linearly ordered set scattered in the case it is scattered under its order topology. For a linearly ordered set L let D L be the set of the Dedekind cuts of L which becomes a linearly ordered set. For instance, D consists of one point and D p consists of two points. We remark that D L becomes compact and 0-dim ensional under its order topology. Theorem 2.2 (Cannon and Conner [1]). Let Sc be the subgroup of i I i consisting of all W s such that W are reduced and D W are scattered. Then Sc is free. Now, we start to prove lemmas for the first theorem. Lemma 2.3. Let G be a free, fine subgroup of Bd n containing Bd n 1 and let A 0 A m be reduced words of appearance n such that i l i A k = n i l i A k =n = for distinct k and k Suppose that A i / G for any 0 i m Then G A i 0 i m = G m i=0 A i holds. Proof. Suppose not. There exist Y 0 Y k = e such that the left term is a non-empty reduced form in G m i=0 A i We assume that k is the minimum in such numbers. Since one of the Y i s is A m or A 1 m we may also assume that Y 0 is A m Now apply Lemma 1.7 to A m = Yk 1 1 Y1 We have reduced subwords W 0 W s of A m and X 0 X s+1 with the properties in Lemma 1.7. If W j is of appearance n W j cannot be a subword of any A i for 0 i<m Since A m / G, G is fine, and G Bd n 1 there exist W j0 and Y k0 such that W j0 is of appearance n and a subword of Yk 1 0 Y k0 = Am or Y k0 = A 1 m Since W j0 is of appearance n and A m is reduced, Yk 1 0 = Am holds and consequently W j0 +1 W j Yk 1 1 0 1 Y1 = X j0 +1W j0 +1 W s X s+1. Therefore, both Yk 1 1 0 1 Y1 = e and Yk 1 1 Yk 0 +1 = e hold, which contradicts the minimality of k The next lemma is well known and easy to prove. Lemma 2.4. Let F be a free group generated by e 1 e n Then F is freely generated also by e 1 e 1 e 2 e 1 e 2 e n

602 katsuya eda Proof of Theorem 2.1. We shall prove that Bd n 1 is a free factor of Bd n and Bd n is free for each n by induction. This is enough to show the theorem. The initial stage: set a given free subgroup G to be Bd 0 i.e., a trivial group. Our induction hypothesis is the following: G is a free, fine subgroup of Bd n and contains Bd n 1 as a free factor. If G = Bd n we have finished the nth step. Otherwise, there exists a reduced word W of appearance n such that W / G Applying Lemma 2.3 for i 0 = m = 0 then we conclude G W = G W We pick initial subwords of W as generators inductively. Suppose that V an initial subword of W freely generates G with G i.e., G =G V V If G = G V an initial subword of W then G V V = G V an initial subword of W is a free, fine subgroup by Lemma 1.6 and we proceed to the procedure to find a reduced word of appearance n outside of the constructed subgroup. Otherwise, there exists an initial subword U of W such that U / G We claim that G U = G V V U To see this, let V 0 V m so that V j is an initial subword of V j+1 for 0 j m 1 and V i is an initial subword of U U being an initial subword of V i+1 and V m = W Let Aj be reduced words such that V 0 for j = 0 V 1 j 1 V j for 0 <j i A j = Vi 1 U for j = i + 1 U 1 V i+1 for j = i + 2 Vj 2 1V j 1 for i + 2 <j m + 1 The facts A j / G for j i and j i + 3 follow from the fact G =G V V The facts A i+1 / G and A i+2 / G follow from U / G V i V i+1 Then, A 0 A m+1 = W holds and hence G Ai 0 i m + 1 = G m+1 i=0 A i by Lemma 2.3. Now, G V i 0 i m U = G m+1 i=0 A i =G m i=0 V i U by Lemma 2.4, which implies the claim. Just iterating the above process transfinitely, we get the desired free generators of Bd n For the proof of Theorem 2.2, we need some definitions. Words V and W are tail-equivalent if there exists a non-empty word X such that V = YX and W = ZX for some words Y and Z Let Sc α be the subgroup of Sc generated by all W s such that W s are reduced and D W α = Then, Sc 1 is the canonical free subgroup consisting of all words of finite length. For an ordinal β 1, let 0 β be the set of all reduced words W such that D W β is a singleton consisting of the largest element of D W Choose a representative from each tailequivalent class of 0 β and let 1 β be the set of such representatives.

free subgroups 603 For V 1 β, anessential part of V is a tail part of V, i.e., a non-empty subword W of V such that V = XW for some W and an essential part of V 1 is a head part of V 1, i.e., a non-empty subword W of V 1 such that V 1 = WX for some W. The next lemma is straightforward from Proposition 1.3, so we omit the proof. Lemma 2.5. Let W be a reduced word with W Sc β and V 1. Then VW and WV 1 are reduced words. Lemma 2.6. Let W be a non-empty reduced word with W Sc β, V 0 V 1 1 β, and ε 0 ε 1 =±1. IfX is the reduced word of V ε 0 0 WVε 1 1, then the essential parts of V ε 0 0 and V ε 1 1 remain in W. Proof. In the case ε 0 ε 1 = 1, one of V ε 0 0 W and WVε 1 1 is reduced and we conclude the essential part of V ε 0 0 and V ε 1 1 remains in X. In the cases ε 0 = 1 and ε 1 = 1, V 0 WV 1 is reduced by Proposition 1.3 and the conclusion is clear. In the remaining cases, i.e., ε 0 = 1 and ε 1 = 1, the reduced word V of WV 1 belongs to 0 β holds. Suppose that the essential part of V is canceled in the reduced word of V0 1 V. Then V 0 and V are tail-equivalent by the fact V 0 V 0 β and Proposition 1.3. Since V and V 1 are tailequivalent, V 0 and V 1 are same. Since V0 1 WV 1 e, the tail part and the head part of V0 1 WV 1 e remain in reduced word V0 1 WV 1 and hence the conclusion holds. Lemma 2.7. Let W be the reduced word of W 0 W 1 W n, where e W i Sc β or W i 1 β V 1 V 1 β for each i; W i Sc β if and only if W i+1 1 β V 1 V 1 β for each 0 i n 1. Then the essential parts of W i 1 β V 1 V 1 β remain in W. Proof. We prove this by induction on n. LetX be the reduced word of W 1 W n. In the case W 0 Sc β, the essential part of W 1 remains in X by the induction hypothesis. Since the essential part of W 1 remains in the reduced word of W 0 W 1, we get the conclusion by Proposition 1.3. In the other case, i.e., W 0 1 β V 1 V 1 β, the essential part of W 2 remains in X and also in the reduced word of W 0 W 1 W 2 by Lemma 2.6. Therefore, the conclusion follows from it. Proof of Theorem 2.2. By induction on α 1 we prove that Sc β is a free factor of Sc α for 1 β<αand Sc α is free, which implies the theorem. Let α be a limit ordinal. By induction hypothesis, we choose a free subgroup T γ of Sc γ+1 so that Sc γ+1 = Sc γ T γ holds. Since Sc γ = δ<γ Sc δ for a limit γ, Sc α = Sc 1 1 γ<α T γ holds and we get the conclusion. In the case α is

604 katsuya eda β + 1 it suffices to show that Sc β is a free factor of Sc α and Sc α is free. Let L be a scattered, compact, linearly ordered set such that L β and L α = Then L β is a finite set. It is easy to see that Sc α is generated by W W 0 β. Suppose that words V and W in 0 β are tailequivalent. There exist a non-empty word X and words Y and Z such that V YX and W ZX by definition. Then Y Z Sc β and hence V Sc β W Therefore, Sc α is generated by Sc β and W W 1 β. Now, Sc α = Sc β W 1 β W by Lemma 2.7. Remark 2.8. Here, we note a counterpart of Theorem 2.2. Let S be the subset of i I i consisting of all W s such that W is an empty word or W are reduced and D W is a perfect set, i.e., compact and without isolated points. S is not a subgroup, but S contains a subgroup which is isomorphic to ω. To see these, let X n n<ω be a partition of ω by infinite subsets, let W n =, and let W n X n be a bijection for each n<ω. Then the natural endomorphism defined by ϕ δ n = W n for each n<ωis injective and maps into S. Since W 0 δ 0 W 0 is reduced and W 0 δ 0 W 0 is isomorphic to, W 0 δ 0 W 0 S and so S is not a group. APPENDIX We explain how elements of the free σ-product correspond to loops in the Hawaiian earring = n=1 x y x 1/n 2 + y 2 = 1/n 2 and under this correspondence what loops correspond to elements of Bd or Sc in case I =. (We refer the reader to [3, Appendix] for a precise definition of this correspondence.) Here, we mean, a loop is a loop in whose base point is 0 0. A loop which winds clockwise to the ith circle corresponds to i and a loop which winds anti-clockwise to the ith circle corresponds to i. LetC be the Cantor ternary set and let n=1 a n b n = 0 1 \C, where a m b m a n b n = for m n. The family a n b n n is naturally ordered according to the order of the real line and its order type is isomorphic to the order of the rationals. Since any countable linearly ordered set can be embedded into the order of the rationals, we first embed W to the ordering of a n b n n. For a word W, we patch windings or the constant map to each a n b n according to W and the embedding of W, and extend the partial map to be a map by mapping the undefined part to 0 0. Then, the resulting map becomes continuous and is a loop corresponding to W. Now, under the above correspondence an element of Bd corresponds to a loop time whose windings, clockwise or anti-clockwise, to each circle are uniformly bounded. On the other hand, loops corresponding to elements of Sc are constructed in a very different way, as follows. We use two types

free subgroups 605 of patching infinitely many loops. For loops f n n, we realize f n in 1/ n + 1 1/n in the first type and f n in 1 1/n 1 1/ n + 1 in the second. The only restriction of these patchings is that the images of f n s converge to 0 0, which assures the resulting map is continuous. Starting from a clockwise winding to a circle, an anti-clockwise winding, or the constant map for f n s, we inductively construct loops using the two types of patchings. These are loops corresponding to elements in Sc. Finally, we remark that a complicated loop may be homotopic to the constant loop. We construct a null homotopic loop which does not belong to the loops introduced for Bd and Sc in the above. For ε i =0 2 4, we realize the clockwise winding to the nth circle in n 1 i=1 ε i /5i + 1/5 n n 1 i=1 ε i /5i + 2/5 n and the anti-clockwise winding to the nth circle in n 1 i=1 ε i /5i + 3/5 n n 1 i=1 ε i /5i + 4/5 n, and map the undefined part to 0 0. Then, times of windings to the nth circle are 2 3 n 1 and so times are not uniformly bounded. Since the order type of the set of the intervals n 1 i=1 ε i /5i + 1/5 n n 1 i=1 ε i /5i + 2/5 n and n 1 i=1 ε i /5i + 3/5 n n 1 i=1 ε i /5i + 4/5 n are the same order type as that of and so the Dedekind cuts form a perfect set. REFERENCES 1. J. W. Cannon and G. R. Conner, The combinatorial structure of the Hawaiian Earring Group, preprint. 2. K. Eda, The first integral singular homology groups of one point unions, Quart. J. Math. Oxford 42 (1991), 443 456. 3. K. Eda, Free σ-products and noncommutatively slender groups, J. Algebra 148 (1992), 243 263. 4. K. Eda, Free σ-products and fundamental groups of subspaces of the plane, Topology Appl. 84 (1998), 283 306. 5. G. Higman, Unrestricted free products, and variety of topological groups, J. London Math. Soc. 27 (1952), 73 81. 6. G. Nöbeling, Verallgemeinerung eines satzes von herrn e. specker, Invent. Math. 6 (1968), 41 55. 7. E. Specker, Additive gruppen von folgen ganzer zahlen, Portugal. Math. 9 (1950), 131 140. 8. A. Zastrow, The non-abelian specker-group is free, preprint.